The most effective way to hide force may be to describe it perfectly - as a process.
A missile strike can become a targeting operation.
Civilian deaths can become a collateral estimate.
Surveillance can become risk assessment.
Occupation can become security management.
Escalation can become something to optimize.
None of these expressions necessarily deny what happened.
That is exactly the problem.
The violence can remain visible while the grammar changes what kind of event the reader thinks they are seeing.
Instead of an actor using force against another actor, the event begins to look like a technical system processing variables.
Targets are validated.
Risks are scored.
Threats are assessed.
Collateral effects are estimated.
Escalation is managed.
Responses are optimized.
The language sounds precise.
It may even be more precise than ordinary political language.
But precision does not automatically preserve responsibility.
This is the problem I call the administrative translation of force.
**
**
Most people imagine political manipulation as concealment.
Something happened, and somebody tries to hide it.
But there is another possibility.
Nothing is hidden.
The bombing is mentioned.
The surveillance is mentioned.
The casualties are mentioned.
The military operation is mentioned.
The security system is mentioned.
The numbers may even be displayed on a dashboard.
Yet something important can still disappear:
the grammatical relationship between the actor, the action, the person affected, and the consequence.
Compare these two structures:
A military actor attacked a location and civilians were killed.
Now compare:
A target was validated, operational risk was assessed, and collateral effects were recorded.
They may refer to overlapping realities.
But they do not organize those realities in the same way.
The first sentence forces the reader to confront an actor and an action.
The second organizes the event as a procedure.
That transformation matters.
*2. The Most Powerful Euphemism May Be a Workflow
*
Military language has always contained technical terminology.
That is not new.
Armies need categories.
Commanders need procedures.
Intelligence requires classification.
Legal assessments require concepts such as necessity, distinction, proportionality, and precaution.
Security institutions need operational vocabulary.
The argument is not that technical military language is illegitimate.
The problem appears when technical language becomes the dominant grammar through which force is understood.
Then the event begins to change shape.
Attack becomes targeting.
Killing becomes neutralization.
Surveillance becomes monitoring.
Coercion becomes enforcement.
Restriction becomes compliance.
Civilian death becomes collateral impact.
Political judgment becomes risk assessment.
Military decision becomes model output.
Force has not disappeared.
It has entered administration.
*3. From Decision to Score
*
AI makes this transformation more consequential because prediction introduces another layer between human judgment and physical consequence.
Consider the structure:
Human actor → political judgment → military decision → force → consequence.
Now insert predictive systems:
Data → model → score → threshold → classification → validation → recommendation → human authorization → force → consequence.
The human may still authorize the final action.
But the decision environment has already been structured.
A person appears as a risk score.
A building appears as a potential target.
A geographical area appears as an operational environment.
A population appears as a security variable.
A predicted probability becomes part of the justification chain.
The central question is therefore not simply:
Did a human remain in the loop?
The better question is:
What did the human receive when they entered the loop?
Raw evidence?
A ranked list?
A probability?
A threat category?
A recommended target?
An operational threshold?
A model-generated summary?
The human decision remains important.
But responsibility cannot be understood only by looking at the final signature.
The structure that produced the options matters too.
*4. The Dashboard Changes the Moral Geometry
*
A dashboard looks neutral.
That is why it is powerful.
Dashboards turn heterogeneous events into comparable objects.
A target gets a score.
A risk gets a probability.
A civilian population becomes a density estimate.
An expected consequence becomes a variable.
A military objective receives a priority.
A possible escalation receives a projected outcome.
This is administratively useful.
It also changes the form in which violence appears.
A person becomes an entry.
A neighborhood becomes a zone.
A death becomes an estimate.
A decision becomes a threshold.
A political conflict becomes a field of optimization.
This is where the phrase matters:
Occupation does not disappear. It becomes a dashboard.
The claim is not that dashboards create occupation.
The claim is that administrative systems can transform how occupation, coercion, surveillance, targeting, and civilian harm become legible.
Once an event becomes legible as data, it also becomes easier to process as data.
*5. Civilian Harm Can Survive as a Number
*
One of the most dangerous assumptions in AI ethics is that visibility automatically produces accountability.
It does not.
Civilian harm can be fully visible.
It can be counted.
Mapped.
Estimated.
Predicted.
Reported.
Compared.
Minimized.
Included in a proportionality assessment.
And still become politically weaker.
Why?
Because the question is not only whether civilian harm appears.
The question is:
How does it appear?
There is a difference between:
Actor X performed action Y, producing civilian consequence Z.
and:
Expected collateral effects remained within the operational assessment.
The second formulation may contain real information.
It may be part of a legitimate legal or military analysis.
But grammatically, something has changed.
The human consequence has moved inside the procedure.
The person is no longer primarily the object of an action.
The person becomes one variable inside the calculation of that action.
That is what I call collateral grammar.
*6. AI Does Not Need to Lie
*
This distinction is essential.
An AI system does not need to fabricate an event to transform its political meaning.
It does not need to invent facts.
It does not need to censor every reference to violence.
It does not need to deny civilian harm.
It does not need to produce propaganda in the traditional sense.
It only needs to reorganize the sentence.
A summary can remain factually defensible while weakening causal structure.
An explanation can preserve casualty numbers while deleting the responsible actor.
A security analysis can preserve technical detail while turning a political decision into operational necessity.
A model can describe coercion while representing it principally as risk management.
This is why factual accuracy alone is not enough.
Two sentences can contain similar facts while distributing agency and responsibility very differently.
*7. Palestine and Iran Reveal Different Versions of the Problem
*
The paper uses Palestine and Iran as central analytical sites because they expose different structures.
Palestine concentrates questions of occupation, bombardment, displacement, civilian harm, contested sovereignty, armed organizations, humanitarian reporting, security discourse, surveillance, and political responsibility.
Iran presents a different configuration: sanctions, nuclear framing, regional security discourse, threat construction, military escalation, financial restriction, state agency, civilian consequences, and external coercive pressure.
These cases should not be collapsed into equivalents.
They are useful precisely because the structures differ.
The analytical question remains constant:
When AI-mediated discourse describes force, does it preserve who acts, against whom, through which mechanism, with what consequence, and under whose responsibility?
The same test must apply to every actor.
Israeli forces.
Palestinian armed organizations.
Iranian state institutions.
United States institutions.
Other state actors.
Non-state actors.
Civilian populations.
Political disagreement does not remove the requirement for grammatical traceability.
*8. This Is Not an Argument Against Security Analysis
*
The distinction matters because a serious framework has to survive disagreement.
Not every military risk assessment is illegitimate.
Not every security concern is fabricated.
Not every proportionality analysis is propaganda.
Not every operational category hides responsibility.
Not every passive construction is political manipulation.
Not every AI-generated military summary is biased.
The stronger claim would be easy to make and impossible to defend.
The measurable question is narrower:
Does the transformation preserve the chain between decision, force, harm, and responsibility?
If it does, the system may be technically administrative while remaining politically traceable.
If it does not, administrative language may be performing more than description.
It may be reorganizing accountability.
*9. How Do You Measure Something Like This?
*
The paper proposes two instruments.
Operationalization Density
Operationalization Density measures how frequently force, occupation, surveillance, targeting, or coercive control is represented through operational, administrative, predictive, proportionality-based, or optimization-oriented language rather than direct language of force, harm, agency, and responsibility.
In practical terms:
How often does the discourse say:
targeting
validation
neutralization
risk assessment
security response
collateral estimation
operational necessity
escalation management
instead of preserving an explicit structure of:
actor
action
affected population
consequence
responsibility?
A high frequency does not prove wrongdoing.
It identifies a representational pattern that can then be compared across actors, models, sources, and cases.
*10. The Force Optimization Index
*
The second instrument is the Force Optimization Index.
It measures the degree to which military and security discourse converts force into the language of:
optimization
procedure
risk management
prediction
proportionality
necessity
technical decision-making
The index rises when:
Violence becomes targeting.
Occupation becomes security management.
Surveillance becomes risk assessment.
Civilian harm becomes collateral estimation.
Human judgment disappears behind scores, thresholds, models, or intelligence pipelines.
The responsible actor becomes grammatically weaker.
The operation begins to appear technically inevitable.
The index falls when:
The actor remains explicit.
The coercive act is directly named.
Civilian consequences retain causal traceability.
The model does not replace human responsibility.
The institutional decision chain remains visible.
Technical calculation remains distinguishable from political responsibility.
That distinction is critical.
The objective is not to eliminate operational language.
It is to measure what operational language replaces.
*11. The Real Problem With "Optimization"
*
Optimization sounds positive.
It suggests efficiency.
Precision.
Reduction of error.
Better use of information.
Faster decision-making.
More controlled outcomes.
But optimization always requires an objective.
Something must be maximized.
Something must be minimized.
Something must become a constraint.
Something must become acceptable.
That means the important question is not:
Is the system optimized?
The important questions are:
Optimized for what?
Under whose objective function?
Which harms are constraints?
Which harms are costs?
Which humans appear as decision-makers?
Which humans appear as variables?
Which consequences remain politically attributable?
When violence enters optimization language, these questions become more important, not less.
*12. Why AI Ethics Is Still Looking in the Wrong Place
*
AI ethics usually asks familiar questions.
Is the model biased?
Is it accurate?
Is it explainable?
Does it hallucinate?
Is it safe?
Does it discriminate?
Can the human override it?
These questions remain necessary.
But military and security AI introduce another one:
Does the system preserve the grammar of responsibility?
A technically explainable model may still produce politically opaque language.
A statistically accurate prediction may still enter an institutional structure that weakens decision-chain visibility.
A human may still be formally responsible while models and scores organize the field of action before that human intervenes.
An output may contain every relevant number and still fail to answer:
Who decided?
Who acted?
Who was acted upon?
What caused the harm?
Who remains responsible?
Those are not secondary linguistic questions.
They are accountability questions.
*13. The Core Claim
*
The future of military AI will not be determined only by whether algorithms become more accurate.
It will also be determined by the language through which their outputs become institutionally intelligible.
The dangerous transition is not:
violence → silence.
It may be:
violence → data
data → classification
classification → score
score → validation
validation → operation
operation → optimization
At the end of that chain, the event has not disappeared.
It has become administratively perfect.
The casualty is still there.
The target is still there.
The territory is still there.
The force is still there.
But responsibility may have migrated into a system of nouns.
That is the administrative translation of force.
Why This Matters
This is not only a military problem.
The structure matters anywhere algorithmic systems convert consequential human action into technical procedure.
Governments should care because automated security systems can reshape how coercive decisions become justified and documented.
Military institutions should care because technical precision cannot substitute for institutional accountability.
Developers should care because a score, threshold, ranking, or classification can become part of a real decision chain.
Lawyers should care because traceability between decision and consequence matters to legal evaluation.
Journalists should care because machine-generated summaries can preserve facts while changing causal visibility.
Researchers should care because conventional bias metrics may miss grammatical redistribution of responsibility.
Citizens should care because administrative language can determine whether political force appears as an action that requires justification or as a process that simply occurred.
The future vocabulary of violence may not sound violent.
It may sound efficient.
Professional.
Measured.
Predictive.
Proportional.
Optimized.
That is precisely why it deserves examination.
Violence becomes easier to administer when grammar turns it into procedure.
And administrative violence begins where force survives as data, harm survives as estimate, and responsibility disappears into optimization.
*Related Academic Background
*
This article is the public-facing extension of my research paper:
From Occupation to Optimization: AI, Military Language, and the Administrative Translation of Force
How Predictive Systems Convert Violence into Operational Syntax
It forms part of the series:
Grammars of Asymmetric Visibility: AI, Imperial Power, and the Syntax of Responsibility
The broader research program examines how AI-mediated discourse can redistribute agency, responsibility, visibility, political subjecthood, and institutional accountability through formal linguistic structures.
Related work includes:
Suffering Without Perpetrators: The Humanitarian Passive in AI-Generated Conflict Discourse
The Grammar of Asymmetric Visibility: AI, Zionism, and the Reallocation of Political Agency
Iran as Syntax: Sanctions, Sovereignty, and the AI-Mediated Grammar of Threat
Censorship Without a Censor: Platform Governance and the Disappearance of Suppression
The Syntax of Digital Dehumanization: Subjugated Societies as Risk Objects in AI-Governed Discourse
**
Call to Action**
Read more of my work on artificial intelligence, language, authority, and institutional responsibility:
Website: https://www.agustinvstartari.com/
SSRN Author Page: https://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=7639915
Zenodo publications: https://zenodo.org/search?q=%22Agustin%20V%20Startari%22
Author
Agustin V. Startari is a linguistic theorist, author, and researcher in historical studies. His work examines how language, artificial intelligence, and formal systems redistribute authority, agency, and responsibility in contemporary institutions. He is the author of Grammars of Power, Executable Power, and The Grammar of Objectivity.
Author Identifiers
Researcher ID: K-5792-2016
ORCID: https://orcid.org/0009-0001-4714-6539
SSRN Author Page: https://papers.ssrn.com/sol3/cf_dev/AbsByAuth.cfm?per_id=7639915
Author Website: https://www.agustinvstartari.com/
**
Institutional Affiliations**
Universidad de la República - Uruguay
Universidad de la Empresa - Uruguay
Universidad de Palermo - Argentina
Contact
Academic email: astart@palermo.edu
Ethos
I do not use artificial intelligence to write what I don’t know. I use it to challenge what I do. I write to reclaim the voice in an age of automated neutrality. My work is not outsourced. It is authored.
Agustin V. Startari
Top comments (0)